categories={'aeroplane', 'bicycle','bird','boat','bottle','bus','car','cat','chair','cow','diningtable','dog','horse','motorbike','person','pottedplant', 'sheep','sofa','train','tvmonitor'};

categs_to_train=[1:20];

year='2012';

cachedir='/work5/bharath2/defdict/cachedir/';

name='defdict';

conf=voc_config;

VOCinit;

ids = textread(sprintf(VOCopts.imgsetpath, 'val'), '%s');

for k=categs_to_train

cls=categories{k};

if(exist(fullfile(cachedir, cls, name, 'ap.mat'), 'file'))

x1=load(fullfile(cachedir, cls, name, 'ap.mat'), 'ap');

aps(k)=x1.ap;

continue;

end

if(~exist(fullfile(cachedir, cls, name), 'file'))

mkdir(fullfile(cachedir, cls, name));

end

if(exist(fullfile(cachedir, cls, name, 'posfeats.mat'), 'file'))

x1 = load(fullfile(cachedir, cls, name, 'posfeats.mat'), 'posboxes', 'feats','featsd');

posboxes = x1.posboxes;

feats = x1.feats;

featsd = x1.featsd;

else

%get the data

[pos, neg, impos] = pascal_data(cls, year);

%use DPM V5 to train a root-only model

mixmodel=pascal_train_rootmodel(cls,3);

%get positive features

[posboxes, feats, featsd]=get_pos_feats(impos, mixmodel);

save(fullfile(cachedir, cls, name, 'posfeats.mat'), 'posboxes', 'feats','featsd');

end

if(exist(fullfile(cachedir, cls, name, 'deformations.mat'), 'file'))

x1=load(fullfile(cachedir, cls, name, 'deformations.mat'), 'destall', 'wtsall', 'lda_weights');

destall = x1.destall;

wtsall = x1.wtsall;

lda_weights = x1.lda_weights;

else

%get deformation dictionaries

[destall, wtsall, lda_weights] = get_deformation_dictionaries(featsd, posboxes);

save(fullfile(cachedir, cls, name, 'deformations.mat'), 'destall', 'wtsall', 'lda_weights');

end

if(exist(fullfile(cachedir,cls, name, 'initmodel.mat'), 'file'))

x1 = load(fullfile(cachedir, cls, name, 'initmodel.mat'), 'model');

model = x1.model;

else

%initialize model

model = init_model_from_def_dict(destall, wtsall, lda_weights, mixmodel);

save(fullfile(cachedir, cls, name, 'initmodel.mat'), 'model');

end

%do the training

model=train_v2_top_cont(cls, name, cachedir, model);

%do the testing

if(exist(fullfile(cachedir, cls, name, 'boxes.mat'), 'file'))

x1 = load(fullfile(cachedir, cls, name, 'boxes.mat'), 'boxes');

boxes = x1.boxes;

else

boxes = model_test_fast_split(name, model, ids, VOCopts);

save(fullfile(cachedir, cls, name, 'boxes.mat'), 'boxes');

end

[ap, prec, recall] = pascal_eval(cls, boxes, 'val', '2012', '');

save(fullfile(cachedir, cls, name, 'ap.mat'), 'ap','prec','recall');

aps(k)=ap;

disp(k);